Tracing signs

From philosophy to linguistics via data science
(and back)

Joshua Wilson Black

Te Kāhui Roro Reo | New Zealand Institute of Language, Brain and Behaviour
Te Whare Wānanga o Waitaha | University of Canterbury

Overview

Philosophical orientation
Data science methods
Two current research streams:
1. sociolinguistics, and
2. experimental history of philosophy
Future directions

You’ve seen me give talks on a series of different projects.
different from my PhD research.
Today: take opportunity to zoom out a little bit.

My doctorate is in philosophy. I’ll start by highlighting a few ideas from my work in philosophy which orient my work. I won’t be arguing for them in this presentation, but am happy to defend them in the question time.
I added a series of ‘data science methods’ to my tool kit, which had previously consisted of ‘read old books and write about them’. So I’ll say a little about why I find them useful and my approach to data science methods in research.
I then talk about my two current research streams, in sociolinguistics and the emerging field of ‘experimental history of philosophy’. I’ll show how they connect with my overall research agenda and some highlights of the work I’ve done while a postdoc at NZILBB.
Finally, I’ll point to some future directions in both of these streams.

Orientation

Philosophical theses

(History of) philosophy: Time depth is essential for understanding anything!
Pragmatism: We are social animals embedded in an environment. No minds without bodies!
Signs: Signs are embodied, come in many forms, and change over time .
Ethics of inquiry: inquiry is essentially communal and generates ethical demands. These include maximal openness regarding both methods and data.

In philosophy there’s a broad division between an approach which starts with ‘problems’ and treats them without any kind of historical consciousness and those which take the history of problems to be important. I’ve always been on the history-is-important side. In particular, I became keen very early on the idea that ‘critical self-understanding’, or the Socratic injunction ‘know thyself’, required an understanding of how you got to where you are, why, and to get a sense of the alternative possibilities.
I because very interested in the work of the American philosopher Charles Peirce, who died around 1914, and the broader tradition called ‘pragmatism’. This is not ‘pragmatism’ in the sense of ‘just do what is most expedient’. Rather, I take the core thought to be something like the following: read.
This leads to a pretty expansive notion of ‘sign’, which is designed to introduce, or at least allow for, a kind of continuity between animal behaviour and the most ‘advanced’ forms of human thought. It is important to emphasise that the relevant environment here consists of social norms as well as the kind of things you can bump in to.
The pragmatist were also lead to a view of inquiry as necessarily social. Peirce didn’t think the notion of truth even made sense in the absence of an ongoing communal practice. This leads to a series of ethical claims about research. If our research is primarily a contribution towards a communal end, then we should be as open as possible (modulo other ethical commitments which we might have to human participants! — the whole picture takes healthy inquiry to depend on healthy sociality, so this isn’t a contradiction).

Motivating question

How can we trace and explain the dynamics of human signs, understood in their broadest terms?

Data science methods

‘Reading old books and writing about them’ has problems:
- scale, sampling, demographics, and narrow ‘semiotic range’
Data science offers new opportunities:
- contemporary computational resources,
- access to large, varied corpora with significant time depth.
Challenges remain (but this is fun!)

Reproduction and Generalisation

Research outputs are:
- part of a communal project, and
- should be designed to build capacity.
Reproducibility is a positive opportunity for sharing techniques and lessons learned.
I aim for the wider goal of methodological generalisability:
- in supplementary materials (Wilson Black 2022b; Wilson Black, Hay, et al. 2022b, 2022a),
- tutorial posts (Wilson Black 2022a, 2022c), and
- R package development (Wilson Black and Brand 2023).
Methodologically generalisable research allows both reproduction and extension.

I want to highlight some key values in this DS work. This is to apply the pragmatist ‘ethics of inquiry’ gestured at earlier.

Reproducibility has been widely discussed as an aspect of the response to the replication crisis in psychology and the other human sciences. To produce reproducible research is to allow others to follow along the whole path from data to results which you report (and other results which might have been important to the process). Don’t underestimate the number of people out there just making up data and plots.

We can do this by simply throwing up whatever script we have used with minimal commentary, or we can take this to be an opportunity to share and explain methods so that they can be taken up by others and in other contexts. This is to produce methodologically generalisable research. [Apologise for name].

This is a value in recent research at NZILBB which I have taken this very seriously: - In supplementary materials with papers, which many of you will know are massive. - In tutorial posts which go into particular methods in more detail than appropriate in supplementary materials (building up plots piece-by-piece for instance), - In R package development (to repackage code from our research in a more easily applicable form).

This is a lot of work!

Stream 1: Sociolinguistics

Tracking vowels

Sociolinguistic variables:
- are signs, and
- draw attention to the embodiment of sign users within a social environment.
Corpora housed at NZILBB offer unique opportunities for tracing change in sociolinguistic variables over the course of a large range of time scales:
- minutes, years, and decades
My research has focused on vowels in terms of:
- Clustering: how vowels function together.
- Acquisition: how vocalic variation is picked up by children.

Sociolinguistic variation is, in the classic definition, different ways of ‘saying’ the same thing which have patterns determined by social context.

Variation of this sort conveys meaning and, in the broad sense, ought to be included within the ambit of ‘signs’. Indeed, Eckert adopting a broadly Peircean notion of ‘indexical’ signs to characterise sociolinguistic variables.

For the pragmatist, these signs are particularly interesting as the foreground the position of the organism within their social and physical environment. for the data scientist, they are appealing, or at least, the variables I will be talking about are appealing because they can be measured.

NZILBB has corpora which allow for this kind of sign to be traced at a large range of time scales, including minutes, years, and decades, each of which I’ll illustrate with research which I have worked on while here.

Minutes

RQ: Do vowels function together as features of styles in the course of a monologue?
Corpus: QuakeBox (Clark et al. 2016; Walsh et al. 2013)
Method: divide monologue into intervals → apply Principal Component Analysis (PCA) → explain results using GAMMs.
Answer: Apparently not!
Unexpected discovery: we found an (overlooked!) influence of amplitude on the vowel space.
Data science methods are powerful exploratory tools.
- The RQ and unexpected discovery differ.
- Our write up is explicit about this structure.
Forthcoming in Linguistics Vanguard (Wilson Black et al. forthcoming)

Years

RQ: How do children acquire and advance existing sound changes?
RQ: Does this match the standard account (Labov 2001)?
- Labov assumes transmission of a female parent’s production to their children.
Corpus: a longitudinal Christchurch preschool story retell corpus (3;11 - 5;5 years).
Data challenges:
- forced alignment (see Fromont et al. Forthcoming),
- formant tracking, and
- modelling and normalisation with limited data.
Result: children start from a more ‘conservative’ place in the New Zealand short vowel shift than expected on Labov’s account.
Work in progress with Lynn Clark (PI), Robert Fromont, and Maggie Blackwood.

Decades

RQ: How can the methods applied by Brand et al. (2021) be strengthened and generalised?
- The ‘loadings’ used to cluster vowels can be unstable . But how unstable?
Corpora: ONZE, QB1, and QB2.
Solution:
- calculate error bars via bootstrapping, and
- compare with null distribution generated by permutation.
Implementation: an R package (nzilbb.vowels) which provides functions to perform the relevant calculations and (Wilson Black and Brand 2023).
Published in Language and Linguistics Compass (2022)

Stream 2: Experimental history of philosophy

An emerging field

Experimental history of philosophy is ‘experimental’ relative to philosophy
- i.e. it includes corpus work.
Current work uses:
- corpus analysis and bibliometrics to track history of key terms and networks of influence (e.g. Betti et al. 2019), and
- semantic networks to identify gaps in current literature (e.g. Alfano 2019).
This is part of a broader shift towards large scale text corpora to investigate human thought (e.g., in psychology, Jackson et al. 2022).

Philosophy in newspapers

My work has looked to English-language New Zealand newspapers up to 1900.
An interesting time:
- public engagement with emerging evolutionary thought,
- establishment of colonial government, and
- world-view interaction and clashes between Māori and Pākehā.
Traditional methods in history of philosophy have found barely any story to tell (e.g. Davies and Helgeby 2014).
Two demographic expansions:
- a geographically remote area, and
- a more diverse population.

Finding philosophy

RQ1: How can philosophical writing be identified in historical newspapers?
- Problem: ‘needle in the haystack’.
RQ2: Can we extract meaningful patterns from extracted texts?
Method:
- Iterative bootstrapping for corpus construction.
- Cooccurrence networks for extracting patterns in use of words.
Results:
- A general method for specialist corpus construction in historical newspapers (Wilson Black 2022d).
- A ‘philosophy’ corpus containing 61,000 items extracted from 7.5 million newspaper articles.
- An interactive for generating semantic networks: https://nznewspaperphilosophy.canterbury.ac.nz/

Current project

A shift from newspapers as text to newspapers as evidence.
Why?: whether a text is ‘philosophy’ is contentious. A less contentious approach is to look directly for evidence of intellectual argument and activity.
Aim: to
- identify advertisements for public debates and lectures (not contentious!),
- extract the who, where, what, and when,
- classify events by content, and
- map them geographically and temporally in a publicly accessible interface.
I am currently producing a pilot study developing these methods using Christchurch newspapers (with seed funding from UC Arts Digital Lab).
These project are unprecedented in scale for applications of data science methods to philosophy.

Future directions

Stream 1

Current pilot study: PCA as a tool for investigating systematic change over the lifespan between QB1 and QB2.
Return to within-speaker stylistic variation.
- Pilot alternative methods for dividing recordings and clustering variables.
Sort out normalisation.
- Widely used methods are insufficiently supported.
- Why do normalisation methods struggle with young children?

Stream 2

Intellectual life in NZ newspapers
- Acquire funding to extend the pilot study of public debates and lectures nationwide.
- Trace the emergence and spread of evolutionary ideas in a ‘bottom-up’ way.
- Build relationships with Māori researchers and capacity in te reo Māori.
Does elite philosophical writing lead or lag public reasoning?
- History-inclined philosophers often attribute changes in society to changes in elite philosophy.
- On the other hand ‘The owl of Minerva spreads its wings only with the coming of the dusk.’
- This relationship is testable in key cases.

Combining streams

Newspaper complaints have been used to track changes in New Zealand English by the ONZE project (Gordon et al. 2004).
NZILBB holds quite a few newspaper clippings of this sort.
Can this approach be scaled?

Tracing signs

I apply contemporary data science methods to large corpora to trace changes in human sign use over multiple time scales.
I aim at both reproducibility and methodological generalisability through extensive documentation and methods sharing in supplementary material, tutorials, and software.

Ngā mihi nui!

References

Alfano, Mark. 2019. Nietzsche’s Moral Psychology. Cambridge University Press. https://doi.org/10.1017/9781139696555.

Betti, Arianna, Hein van den Berg, Yvette Oortwijn, and Caspar Treijtel. 2019. “History of Philosophy in Ones and Zeros.” In Methodological Advances in Experimental Philosophy. Bloomsbury Academic. https://doi.org/10.5040/9781350069022.ch-011.

Black, Joshua. 2013. “Peirce on Habit, Practice, and Theory: The Priority of Practice and the Autonomy of Theory.” Master’s thesis, University of Waikato.

———. 2017. “Peirce’s Conception of Metaphysics.” PhD thesis, University of Sheffield.

Brand, James, Jen Hay, Lynn Clark, Kevin Watson, and Márton Sóskuthy. 2021. “Systematic Co-Variation of Monophthongs Across Speakers of New Zealand English.” Journal of Phonetics 88 (September): 101096. https://doi.org/10.1016/j.wocn.2021.101096.

Clark, Lynn, Helen MacGougan, Jennifer Hay, and Liam Walsh. 2016. “‘Kia Ora. This Is My Earthquake Story.’ Multiple Applications of a Sociolinguistic Corpus.” Ampersand 3: 13–20. https://doi.org/10.1016/j.amper.2016.01.001.

Davies, Martin, and Stein Helgeby. 2014. “Idealist Origins: 1920S and Before.” In History of Philosophy in Australia and New Zealand, edited by Graham Oppy and Nick Trakakis, 15–54. Springer.

Fromont, Robert, Lynn Clark, Joshua Wilson Black, and Margaret Blackwood. Forthcoming. “Maximizing Accuracy of Forced Alignment for Spontaneous Child Speech.” Language Development Research, Forthcoming.

Gordon, Elizabeth, Campbell Lyle, Hay Jennifer, Maclagan Margaret, Sudbury, and Peter Trudgill. 2004. New Zealand English: Its Origins and Evolution. Cambridge University Press.

Jackson, Joshua Conrad, Joseph Watts, Johann-Mattis List, Curtis Puryear, Ryan Drabble, and Kristen A. Lindquist. 2022. “From Text to Thought: How Analyzing Language Can Advance Psychological Science.” Perspectives on Psychological Science 17 (3): 805–26. https://doi.org/10.1177/17456916211004899.

Labov, William. 2001. Principles of Linguistic Change: Social Factors. Wiley-Blackwell.

Legg, Catherine, and Joshua Black. 2020. “What Is Intelligence for? A Peircean Pragmatist Response to the Knowing-How, Knowing-That Debate.” Erkenntnis 87 (5): 2265–84. https://doi.org/10.1007/s10670-020-00301-9.

Walsh, Liam, Jen Hay, Derek Bent, Jeanette King, Paul Millar, Viktoria Papp, and Kevin Watson. 2013. “The UC QuakeBox Project: Creation of a Community-Focused Research Archive.” New Zealand English Journal 27: 20–32.

Wilson Black, Joshua. 2022a. “Model Check and Significance Testing for Vowel Space GAMMs.” 2022. https://joshua.wilsonblack.nz/post/significance-testing-for-vowel-space-gamms/.

———. 2022b. “Supplementary Material for "Creating Specialized Corpora from Digitized Historical Newspaper Archives".” 2022. https://doi.org/10.17605/OSF.IO/7CRGT.

———. 2022c. “Visualising Vowel Space Change with GAMMs.” Linguistics Methods Hub. October 28, 2022. https://doi.org/10.5281/zenodo.7261966.

———. 2022d. “Creating Specialized Corpora from Digitized Historical Newspaper Archives.” Digital Scholarship in the Humanities 38 (2): 779–97. https://doi.org/10.1093/llc/fqac079.

———. 2023. “Peirce on Metaphysics and Commonsense Belief.” In Pragmatic Reason, 195–210. Routledge. https://doi.org/10.4324/9781003165699-13.

Wilson Black, Joshua, and James Brand. 2023. “Nzilbb.vowels: Functions for Vowel Covariation Studies (0.1.0-Alpha).” https://doi.org/10.5281/zenodo.8303224.

Wilson Black, Joshua, James Brand, Jen Hay, and Lynn Clark. 2022. “Using Principal Component Analysis to Explore Co-Variation of Vowels.” Language and Linguistics Compass 17 (1). https://doi.org/10.1111/lnc3.12479.

Wilson Black, Joshua, Jen Hay, Lynn Clark, and James Brand. forthcoming. “The Overlooked Effect of Amplitude on Within-Speaker Vowel Variation.” Linguistics Vanguard, forthcoming.

———. 2022a. “Supplementary Material for "the Overlooked Effect of Amplitude on Within-Speaker Vowel Covariation".” 2022. https://nzilbb.github.io/amp_f1_public/.

———. 2022b. “Supplementary Material for "Using Principal Component Analysis to Explore Co-Variation of Vowels".” 2022. https://nzilbb.github.io/PCA_method_supplementary/pca_method_supplementary.html .